GANs can be made fully convolutional so that you can generate images of abritrary size. But the training of the latent space will still be limited by memory and

GANs can be made fully convolutional so that you can generate images of abritrary size. But the training of the latent space will still be limited by memory and speed, so you'd end up with images that are perhaps quite repetetive in nature.